What question did this study set out to answer?

This research aims to establish a framework for continuous admissibility in AI governance, addressing gaps in existing regulatory structures.

July 7, 2026Open Access

Continuous Admissibility: A Temporal Governance Framework for Delegated Machine Authority in Regulated Enterprises

Key Points

This research aims to establish a framework for continuous admissibility in AI governance, addressing gaps in existing regulatory structures.
Introduced the Continuous Admissibility Model with three components: Continuous Admissibility Monitor, ADAE Reassessment Obligation, and Structural Change Trigger.
Conducted a survey of major existing governance frameworks to identify their shortcomings regarding continuous admissibility.
Proposed theoretical framing for admissibility decay based on changing operational conditions.
Existing frameworks like ISO/IEC 42001 and the EU AI Act do not structurally address continuous admissibility.
Identified the need for a mechanism to ensure ongoing validation of machine authority under varying conditions.
Demonstrated implications for regulatory frameworks regarding machine authority management and delegation.

Abstract

Existing AI governance frameworks govern the moment of deployment. Systems are authorised, assessed, and approved before they go live. What happens between deployment and the next scheduled review cycle is structurally ungoverned: no mechanism exists in ISO/IEC 42001:2023, the EU AI Act, or APRA CPS 230 to test whether the conditions under which a system was authorised still hold, or whether a delegation that was correctly granted at the time of initial authorisation remains admissible under current operating conditions. This paper formalises a governance dimension that existing frameworks do not reach: continuous admissibility. It establishes the conceptual distinction between historical validity, whether machine authority was correctly granted at T0, and current admissibility, whether that authority retains the right to bind consequences at T1. It introduces the Continuous Admissibility Model as an operational governance instrument comprising three components: the Continuous Admissibility Monitor, which detects live operating parameter drift across five defined dimensions spanning both a system’s own operating pattern and the currency of the facts and data sources its authorisation depends on; the Autonomous Decision Authority Exposure (ADAE) Reassessment Obligation, which mandates reassessment when drift is confirmed; and the Structural Change Trigger, which invalidates prior authorisations on material organisational discontinuity. It proposes a theoretical framing for admissibility decay: the rate at which a historically valid delegation loses its current admissibility as conditions change. It demonstrates that the major existing governance frameworks surveyed do not address this dimension structurally and identifies specific implications for ISO/IEC 42001, the EU AI Act, and prudential regulatory frameworks. The paper is a companion to Hossain (2026), which introduced the Autonomy Budget construct for portfolio-level authority governance. Where the Autonomy Budget addresses the spatial dimension of machine authority: how much exists in aggregate across a portfolio, this paper addresses the temporal dimension: whether individual delegations within that portfolio retain the right to bind consequences over time.

Read Full Paperexternally

Ask AI

Helpful

Bookmark

View Full Paper